Application of the collision mathematical model based on a BP neural network in railway vehicles

Abstract

In order to enhance the learning performance of small-data-set models and improve the computation efficiency of finite element simulations of vehicle collision, the collision mathematical model (VCMM) based on the back-propagation (BP) neural network is established to predict the collision response data of a single car and marshalling cars at unknown velocities. The predicted results of VCMM were compared with the simulation results of the finite element method (FEM) to verify the model. The compared results show that the maximum relative errors of deformation, energy absorption and average interfacial force of a single vehicle are all below 8.5%, and the relative errors of the maximum compression of the C0 coupler and the internal energy of the A1 car among the marshalling cars are all less than 5%. In addition, the calculation time of the single car and marshalling cars collisions based on the VCMM are reduced by 24.36 and 61.8 times, respectively, compared with the FEM results, and the simulation calculation efficiency is greatly improved. The prediction result of VCMM will partially replace experimental and simulation results for crashworthiness and safety design of the vehicle structure in future studies.